accurcy problems

[kk7nc]

accurcy problems

[THtanghuan]

Thank you very much!!
But,"x_train = np.empty((num_train_samples, 3, 100, 100), dtype='uint8')" should be true,as I run the code as you said,having the following question:

ValueError: could not broadcast input array from shape (120,3,100,100) into shape (120,100,100,3)

Thanks

[kk7nc]

Thank you for your question,
The could happen in many possible issue, as we explain NOT all models in rmdl will be good but majority votes helped to get good results, as I see in your results the RNN contains acceptable results, probability for DNN and CNN contains good results so you should look at your feature space of your dataset
Do you apply RMDL for text or image dataset?

[THtanghuan]

Thank you for your reply.
Yes,I apply RMDL for classification in my own image dataset(size(100x100)).About this question " val_acc: 0.1000",there has no this problem in MNIST and CIFAR datasets,but in my datasets have this question.I restart the code for many times,there will be no these problem.If there are many " val_acc: 0.1000"s in these models, will it have a great influence in VOTEs??
Thanks!

one:
"""
/home/nchen/anaconda3/bin/python /media/nchen/DISK_E/TH/RMDL-master/Examples/YPT.py
Using TensorFlow backend.
sys.version_info(major=3, minor=6, micro=3, releaselevel='final', serial=0)
sys.version_info(major=3, minor=6, micro=3, releaselevel='final', serial=0)
[nltk_data] Downloading package stopwords to /home/nchen/nltk_data...
[nltk_data] Package stopwords is already up-to-date!
DNN 0

(100, 100, 3)
<keras.optimizers.RMSprop object at 0x7f5e6ff8ff60>
Train on 7000 samples, validate on 3000 samples
Epoch 1/100

• 4s - loss: 14.3419 - acc: 0.0986 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00001: val_acc improved from -inf to 0.10000, saving model to weights\weights_DNN_0.hdf5
Epoch 2/100

• 2s - loss: 14.5063 - acc: 0.1000 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00002: val_acc did not improve from 0.10000
Epoch 3/100

• 2s - loss: 14.5063 - acc: 0.1000 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00003: val_acc did not improve from 0.10000
Epoch 4/100

• 2s - loss: 14.5063 - acc: 0.1000 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00004: val_acc did not improve from 0.10000
Epoch 5/100

"""

two:
"""
/home/nchen/anaconda3/bin/python /media/nchen/DISK_E/TH/RMDL-master/Examples/YPT.py
Using TensorFlow backend.
sys.version_info(major=3, minor=6, micro=3, releaselevel='final', serial=0)
sys.version_info(major=3, minor=6, micro=3, releaselevel='final', serial=0)
[nltk_data] Downloading package stopwords to /home/nchen/nltk_data...
[nltk_data] Package stopwords is already up-to-date!
CNN 0

<keras.optimizers.Adagrad object at 0x7f9c41d60518>
Train on 7000 samples, validate on 3000 samples
Epoch 1/400

• 74s - loss: 14.3458 - acc: 0.0991 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00001: val_acc improved from -inf to 0.10000, saving model to weights\weights_CNN_0.hdf5
Epoch 2/400

• 68s - loss: 14.5063 - acc: 0.1000 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00002: val_acc did not improve from 0.10000
Epoch 3/400
...
"""

three:
"""
/home/nchen/anaconda3/bin/python /media/nchen/DISK_E/TH/RMDL-master/Examples/YPT.py
Using TensorFlow backend.
sys.version_info(major=3, minor=6, micro=3, releaselevel='final', serial=0)
sys.version_info(major=3, minor=6, micro=3, releaselevel='final', serial=0)
[nltk_data] Downloading package stopwords to /home/nchen/nltk_data...
[nltk_data] Package stopwords is already up-to-date!
CNN 0

Error in model 0 try to re-generate another model
CNN 0

<keras.optimizers.RMSprop object at 0x7ff9ed1cf240>
Train on 7000 samples, validate on 3000 samples
Epoch 1/400

• 120s - loss: 14.3388 - acc: 0.0993 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00001: val_acc improved from -inf to 0.10000, saving model to weights\weights_CNN_0.hdf5
Epoch 2/400

• 112s - loss: 14.5063 - acc: 0.1000 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00002: val_acc did not improve from 0.10000
Epoch 3/400

• 113s - loss: 14.5063 - acc: 0.1000 - val_loss: 14.5063 - val_acc: 0.1000

Epoch 00003: val_acc did not improve from 0.10000
Epoch 4/400
...
"""

[kk7nc]

Did you normalize your data before sending it to RMDL/?

[kk7nc]

Can you share your code here?

[THtanghuan]

Thank you very much!
I transfer my images into the format of Cifar-10,and using the loading code of cifar-10.
[####code1:YPT.py
'''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''
RMDL: Random Multimodel Deep Learning for Classification

• Copyright (C) 2018 Kamran Kowsari [email protected]
• Last Update: May 3rd, 2018
• This file is part of RMDL project, University of Virginia.
• Free to use, change, share and distribute source code of RMDL
• Refrenced paper : RMDL: Random Multimodel Deep Learning for Classification
• Refrenced paper : An Improvement of Data Classification using Random Multimodel Deep Learning (RMDL)
• Comments and Error: email: [email protected]
  '''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''''
  import sys
  sys.path.append("/media/nchen/DISK_E/TH/RMDL-master")

from keras.datasets import cifar10
from RMDL import RMDL_Image as RMDL
from load_selfdatasets import yuputu

if name == "main":
number_of_classes = 10
shape = (100, 100, 3) #(64, 64, 1)

# (x_train, y_train), (x_test, y_test) = cifar10.load_data()
(x_train, y_train), (x_test, y_test) = yuputu.load_data()
x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255

batch_size = 100 #100
sparse_categorical = 0
n_epochs = [100, 150, 200]  ## DNN--RNN-CNN   [500, 500, 500]
Random_Deep = [0, 0, 5]  ## DNN--RNN-CNN

RMDL.Image_Classification(x_train, y_train, x_test, y_test,(100,100,3),
                          batch_size=batch_size,random_deep=Random_Deep,
                          epochs=n_epochs)

######code2:yuputu.py
"""CIFAR10 small images classification dataset.
"""
from future import absolute_import
from future import division
from future import print_function

import sys
sys.path.append("/media/nchen/DISK_E/TH/RMDL-master")
from load_selfdatasets.load_batch_YPT import load_batch
from keras import backend as K
import numpy as np
import os

def load_data():
# dirname = 'cifar-10-batches-py'
# origin = 'https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz'
# path = get_file(dirname, origin=origin, untar=True)

# path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-bin'
# path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-green-bin'
path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-pink-bin'

num_train_samples = 7000

x_train = np.empty((num_train_samples, 3, 100, 100), dtype='uint8')
y_train = np.empty((num_train_samples,), dtype='uint8')

for i in range(1, 6):
    fpath = os.path.join(path, 'data_batch_' + str(i))
    (x_train[(i - 1) * 1400: i * 1400, :, :, :],
     y_train[(i - 1) * 1400: i * 1400]) = load_batch(fpath)

fpath = os.path.join(path, 'test_batch')
x_test, y_test = load_batch(fpath)

y_train = np.reshape(y_train, (len(y_train), 1))
y_test = np.reshape(y_test, (len(y_test), 1))

if K.image_data_format() == 'channels_last':
    x_train = x_train.transpose(0, 2, 3, 1)
    x_test = x_test.transpose(0, 2, 3, 1)

return (x_train, y_train), (x_test, y_test)

#####code3:load_batch_YPT.py
from future import absolute_import
from future import division
from future import print_function

import sys
sys.path.append("/media/nchen/DISK_E/TH/RMDL-master")
from load_selfdatasets.load_batch_YPT import load_batch
from keras import backend as K
import numpy as np
import os

def load_data():
# dirname = 'cifar-10-batches-py'
# origin = 'https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz'
# path = get_file(dirname, origin=origin, untar=True)

# path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-bin'
# path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-green-bin'
path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-pink-bin'

num_train_samples = 7000

x_train = np.empty((num_train_samples, 3, 100, 100), dtype='uint8')
y_train = np.empty((num_train_samples,), dtype='uint8')

for i in range(1, 6):
    fpath = os.path.join(path, 'data_batch_' + str(i))
    (x_train[(i - 1) * 1400: i * 1400, :, :, :],
     y_train[(i - 1) * 1400: i * 1400]) = load_batch(fpath)

fpath = os.path.join(path, 'test_batch')
x_test, y_test = load_batch(fpath)

y_train = np.reshape(y_train, (len(y_train), 1))
y_test = np.reshape(y_test, (len(y_test), 1))

if K.image_data_format() == 'channels_last':
    x_train = x_train.transpose(0, 2, 3, 1)
    x_test = x_test.transpose(0, 2, 3, 1)

return (x_train, y_train), (x_test, y_test)

Thanks.

[kk7nc]

x_train = np.empty((num_train_samples, 3, 100, 100), dtype='uint8') should be
x_train = np.empty((num_train_samples, 100, 100, 3), dtype='uint8')

[kk7nc]

try:

def load_data():
   # path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-bin'
   # path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-green-bin'
   path = '/media/nchen/DISK_E/TH/RMDL-master/dataset-100-random-pink-bin'

   num_train_samples = 7000

   x_train = np.empty((num_train_samples, 3, 100, 100), dtype='uint8')
   y_train = np.empty((num_train_samples,), dtype='uint8')


   for i in range(1, 6):
       fpath = os.path.join(path, 'data_batch_' + str(i))
      (x_train[(i - 1) * 1400: i * 1400, :, :, :],
       y_train[(i - 1) * 1400: i * 1400]) = load_batch(fpath)

    fpath = os.path.join(path, 'test_batch')
   x_test, y_test = load_batch(fpath)

  y_train = np.reshape(y_train, (len(y_train), 1))
  y_test = np.reshape(y_test, (len(y_test), 1))

   if K.image_data_format() == 'channels_last':
       x_train = x_train.transpose(0, 2, 3, 1)
       x_test = x_test.transpose(0, 2, 3, 1)


   x_train = np.reshape(x_train.shape[0],100,100,3)
   x_test = np.reshape(x_trest.shape[0],,100,100,3)


    return (x_train, y_train), (x_test, y_test)

and following code for RMDL:

from keras.datasets import mnist
import numpy as np


from RMDL import RMDL_Image as RMDL

if __name__ == "__main__":
    (X_train, y_train), (X_test, y_test) = load_data()
    X_train = X_train_D / 255.0
    X_test = X_test_D / 255.0
    number_of_classes = np.unique(y_train).shape[0]
    shape = (100,100, 3)
    batch_size = 128
    sparse_categorical = 0

    n_epochs = [100, 100, 100]  ## DNN--RNN-CNN
    Random_Deep = [3, 3, 3]  ## DNN--RNN-CNN
    RMDL.Image_Classification(X_train, y_train, X_test, y_test,shape,
                             batch_size=batch_size,
                             sparse_categorical=True,
                             random_deep=Random_Deep,
                             random_optimizor=False,
                             epochs=n_epochs)

[THtanghuan]

Thanks.
Can I set the shape of input image be [100,1000,3] ? (height:100,width:1000,channel:3)

[AlexYoung757]

ah. i have a question. how to set saving path of model

[kk7nc]

ah. i have a question. how to set saving path of model

Thank you for your question, the model is saved next to your file, but if you want to manually set it we did not send a parameter for this version,
Probably we add it in next version,
Best regards,
Kamran